Research map / Label-efficient and robust learning / Recognition and long-tail
Spread Spurious Attribute: Improving Worst-group Accuracy with Spurious Attribute Estimation
Junhyun Nam, Jaehyung Kim, Jaeho Lee, Jinwoo Shin
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SSA with pseudo-attribute estimation and Group DRO | worst-group robust training · Waterbirds | worst-group accuracy: 86.7 -> 89.0 | Improved |
| Use SSA with pseudo-attribute estimation and Group DRO | worst-group robust training · CelebA | worst-group accuracy: 81.1 -> 89.8 | Improved |
| Use SSA with pseudo-attribute estimation and Group DRO | worst-group robust training · MultiNLI | worst-group accuracy: 72.6 -> 76.6 | Improved |
| Use SSA with pseudo-attribute estimation and Group DRO | worst-group robust training · CivilComments-WILDS | worst-group accuracy: 69.3 -> 69.9 | Improved |
| Add fixed-threshold pseudo-labeling (ablation) | spurious-attribute predictor · CelebA, Blond/Male group, 10% validation labels | group accuracy: 70.4 -> 76.4 | Improved |
| Add fixed-threshold pseudo-labeling (ablation) | spurious-attribute predictor · CelebA, Blond/Male group, 5% validation labels | group accuracy: 68.7 -> 67.8 | Hurt |
| Add group-wise adaptive thresholds (ablation) | pseudo-labeling · CelebA, Blond/Male group, 10% validation labels | group accuracy: 76.4 -> 83.6 | Improved |
| Add group-wise adaptive thresholds (ablation) | pseudo-labeling · CelebA, Blond/Male group, 5% validation labels | group accuracy: 67.8 -> 75.7 | Improved |
| Split group-unlabeled samples into predictor-training and prediction subsets (ablation) | SSA pseudo-labeling · CelebA, Blond/Male group, 10% validation labels | group accuracy: 76.3 -> 83.6 | Improved |
| Split group-unlabeled samples into predictor-training and prediction subsets (ablation) | SSA pseudo-labeling · CelebA, Blond/Male group, 5% validation labels | group accuracy: 71.7 -> 75.7 | Improved |
| Split group-unlabeled samples into predictor-training and prediction subsets (ablation) | SSA pseudo-labeling · Waterbirds, Landbird/Land group, 10% validation labels | group accuracy: 91.3 -> 92.1 | Improved |
| Split group-unlabeled samples into predictor-training and prediction subsets (ablation) | SSA pseudo-labeling · Waterbirds, Waterbird/Land group, 10% validation labels | group accuracy: 92.9 -> 94.6 | Improved |
Closest papers by meaning
- Improving Group Robustness on Spurious Correlation Requires Preciser Group Inference (ICML 2024 · similarity 0.90)
- Let Samples Speak: Mitigating Spurious Correlation by Exploiting the Clusterness of Samples (CVPR 2025 · similarity 0.89)
- Mitigating Spurious Correlations via Disagreement Probability (NeurIPS 2024 · similarity 0.86)
- Label-Efficient Group Robustness via Out-of-Distribution Concept Curation (CVPR 2024 · similarity 0.85)
- Spurious Correlations in High Dimensional Regression: The Roles of Regularization, Simplicity Bias and Over-Parameterization (ICML 2025 · similarity 0.84)
- SEBRA : Debiasing through Self-Guided Bias Ranking (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
